Thursday, June 6, 2013

Newspoll Upfront Exclusion Rates Since 2002

Advance Summary

1.  Some observers think that the proportion of voters who are indicated by Newspoll as "uncommitted" or "refused" gives Labor a much greater chance in the upcoming federal election, because the Government may attract a late swing from undecided voters.

2. The view that there is generally a late swing to an incumbent Government is false.

3. The view that there is usually a late swing to the party that is trailing is true, but it is not clear whether this is because there is a genuine "narrowing" effect, or whether this is because the Coalition tends to outperform its polling and the Coalition has often been the trailing party.

4. It is possible (and logical) that there is a link between increases in the Newspoll exclusion rate and poor polling for Labor, suggesting that when the party is performing poorly more of its voters become "uncommitted".  However, the evidence on this is inconclusive.

5. While it is plausible that undecided voters will move back to the Government and improve its result by, say, two points, this is also something that may not happen at all.

6. The issue of uncommitted and refused rates is only relevant to the margin of an overwhelmingly likely Coalition win (assuming no Labor leadership change).  There is nothing to suggest it is capable of turning the election.
------------------------------------------------------------------------------------------------------------------------



As I've commented in a few other recent pieces, Labor's federal polling over the last few months has been dire, and indeed the Government has not been polling competitively since the election date was announced at the end of January.   Yet there are still some who hold out hope that either public opinion will change radically, or that the polls are simply wrong, and who therefore give the Government a more than token chance of survival.  I addressed prospects for a rapid swing in polled public sentiment in a recent article, How Common Is A Five-Point Swing In Under Four Months? and found that there was not a lot to see there by way of chances - sharp swings in favour of governments do sometimes occur, but are not likely to happen specifically in the last few months of a government's term. 

Views that the polls are simply wrong (and that the actual state of public opinion is that Labor is competitive, if perhaps still trailing, now) mainly rely on the following objections:

1. Landline-vs-mobile-phone issues.
2. The high rate of Others voting and the difficulty determining the composition of those Others.
3. The proportion of voters who do not give a useful answer when asked and are therefore excluded from the poll.

I don't think there's any juice to speak of in (1) since it can be accommodated by scaling (with a slightly increased risk of inaccuracy) and since polls that poll mobiles, polls that do not poll mobiles, and polls that poll online are now all getting very similar average results.

I do think (2) might have a bit of nibble to it since there is suggestive evidence that perhaps two points of the Others vote consists of disgruntled moderates who strongly support Kevin Rudd.  How these people will vote (if they bother voting at all) is unpredictable.  Beyond that, right-wing populist parties KAP and now PUP are likely to be contributing to the Others surge.

This article concerns objection (3), specifically as it applies to Newspoll.  I'm using the term "upfront exclusion" in this article as that is the term used by Twitter Newspoll-watcher @OzEquitist, whose relentless focus on this aspect of Newspoll inspired me to have a close look at it.

Notable flashpoints in online mutterings about upfront exclusion have included the July 6-8 2012 Newspoll (8% uncommitted 3% refused) and the Aug 17-19 2012 Newspoll (9% uncommitted 2% refused). As it happens, these were not quite records; 12% were excluded upfront from the 2-4 Dec 2005 Newspoll.  Another interesting case was the December 2012 Newspoll, taken at the time of Tony Abbott's AWU mudslinging.  The refusal rate of 4% had not appeared since April 2007 (the highest refusal rate in my records being 4.5% shortly before the 2004 election).  

The argument is that these high levels of indecision/refusal represent a soft Labor vote that will return to the party as the election date approaches.  Those trying to ground this in any sort of electoral history argue that there is a history of undecided voters getting behind either (1) the Government of the day or (2) the underdog party in polling.

Is there actually such a history?

Considering the elections since 1970, there were three (1972, 1974 and 1996) in which the final 2PP result basically mirrored what (in the first two cases very limited) campaign-month polling I've seen. Of the rest:

The Government of the day significantly outperformed its polling during the last month on a 2PP basis: 1977, 1980, 1983, 1993, 1998, 2004, 2007

The Government of the day underperformed compared to its polling during the last month: 1984, 1987, 1990, 1996, 2001, 2010

1975 can be argued in various ways since Fraser was campaigning as incumbent (albeit caretaker) PM, but his government was not really a known quantity, and since polling during the campaign blew out in the Coalition's favour, but Labor did not do quite as badly as the last pre-election Morgan Gallup. 

It's such an even split that to argue that the position of governments tends to improve in the last month because they are the government is just discredited.

Now here's the other suggested way of splitting them up:

The party trailing during the last month outperformed its last month polling on a 2PP basis: 1980*#, 1983, 1984, 1987, 1990#, 1993*#, 1998*, 2001, 2007, 2010 (* = won election, #= won 2PP)

The party trailing during the last month underperformed its last month polling:  1975, 1977

I've omitted 2004 because the lead changed about midway through the last month.  

Now that looks rather impressive as evidence of an underdog/narrowing effect, but there is an equally well supported alternative explanation:

The Coalition outperformed its last month polling on a 2PP basis: 1977, 1980, 1983, 1984, 1987, 1990, 1998, 2004, 2007, 2010

The ALP outperformed its last month polling on a 2PP basis: 1993, 2001

Is the underdog/narrowing effect the cause or just the historic Labor Fail Factor?  Take your pick!  But if it's the latter, it mirrors what I found in a TT piece last year.  In that piece, I found that incumbency is an advantage in overcoming bad polling through a government's term generally (considering that governments usually win although the long-term 2PP average for them is about 50:50) but that Coalition governments are much better at overcoming the worst polling slumps of their term than Labor ones are. 

Of course, those who believe upfront exclusion will make a difference may not be concerned with history, and more concerned with aspects of the current situation that they think will make a difference.

Newspoll upfront exclusion rates


Newspoll normally excludes two kinds of respondents - those who refuse to answer a question about voting intention, and those who insist that they have no committed view, even when prodded with the option of just giving the party they are leaning to.  It is assumed that these voters will vote in identical proportions to everyone else.  During election leadups a point or two of intending informal voters are also sometimes excluded.

Unfortunately Newspoll doesn't tabulate these rates historically.  However its website does include a number of "table" PDFs for given national polls, including most polls taken in 2002, a fair minority from 2003 and pretty much all from 2004 onwards.  I've used these as my base for the following yearly averages:


 Here it is in graph form:


 The following are some points from these data:

* The last two years have seen the highest levels of upfront exclusion in the last 12 years, but they are not massively greater than 2003 and 2006.

* There seems to be a cyclical aspect - the year before the election in all cases had a higher upfront exclusion rate than the election year or the year after the previous election.

* The uncommitted rate has been higher during 2012 and 2013 than any of the previous ten years.

* In the total upfront exclusion rate, this is masked somewhat by the Refused rate having declined over the period covered (with a slight upturn in the last two years).

* Upfront exclusion rates in the 2008-2010 cycle were much lower than in the other three cycles shown.  This may be partly explained by the Rudd Government being very popular early in its term.

The obvious question to ask here is what happens during election years.  It might be expected, for instance, that the uncommitted rate drops steadily as the election approaches and more and more voters make up their minds.  But in 2009, while researching incorrect claims that Tasmanian pollster EMRS's state-polling "undecided" rates are not unusually high, I found (see this old TT article) that for state-level polling:

 "Although Newspoll state undecided rates fall sharply very close to an election, there is not a clear and consistent pattern of decline through the electoral cycle and as the campaign approaches.  Indeed, in Queensland from Nov 06 up til the start of the 2008 campaign, the pattern, if anything, pointed in the opposite direction: 4, 5, 4, 4, 4, 5, 6, 6, 8 [..]"

So here are the federal upfront exclusion rate patterns for 2004, 2007, 2010.  In each case, the poll marked E is the last before the election.  In the case of 2010 I have marked the point at which Julia Gillard became Prime Minister with a G.






Note that the interval of polls is not always even; they tend to become weekly rather than fortnightly as the election gets very close.  In 2004 the upfront exclusion rate did not decrease as the election approached until the very final poll. (In a sense it increased, though if we assume those responding "informal" were honest then that increase is largely explained.) In 2007 there was a modest decrease over the last six weeks before the election, and in 2010 rates in the last two months before the election were slightly lower than in the first five months of the year.  Generally, though, there is not a big dropoff in the exclusion rate (and especially the Uncommitted rate) until the very last poll in the final week. In the cases of both 2004 and 2010, the average for the year is dragged down by the addition of post-election polling with low uncommitted rates.

It's important that the "uncommitted" rate often doesn't really crash until the last few days of the election, because polling at that time is generally pretty accurate.  Parties have sometimes lost two or three points to very late swings (Labor under Hawke did this routinely, and they also did it under Latham, while the Coalition did it in 2001) but never yet have they lost five.  There isn't any previous case of the undecided voters more or less all making up their mind in the same direction.

Exclusion rates and voting intention

So, we know that there's a higher than normal "uncommitted" rate in Newspoll this year and we'd expect that the upfront exclusion rate of about 9 will fall to about 3 in the final pre-election poll, perhaps with some decline before then.  That's about six points of voter sentiment that's not currently being registered, but that will be included in the polls by the last few days.  The great hope of quite a few online progressives is that nearly all of these are soft Labor voters and that they will return to the fold on election day.  Is there any basis for this hope?

If there is, one would (all else being equal) expect for starters to find that the numbers of upfront-excluded voters move strongly with Labor's position in the polls, so that when Labor is polling badly, there are more voters refusing to express a preference, and when Labor is polling less badly, there are fewer. 

In looking at this, there are a couple of little traps.  The first one is to just take the whole history of polling over this term and compare the exclusion rate with the 2PP since the last election.  But we already know that there is a cyclical tendency for the upfront exclusion rate to be higher in the second year of a term than the first, and also that Labor's polling was much better early in the term.  So there's a risk there of picking up a misleading correlation.  To avoid this, I looked at polling starting from a point at which Labor's polling was already pretty bad and close to the term average, April 2011.

The other trap is that we're comparing a relatively small number that is very much influenced by rounding (the exclusion rate) with a much larger and quite bouncy number (the 2PP).  In these circumstances, it's easy for a relationship to be hidden in the noise.  So to try to strip out some of the noise I've thrown in a degree of smoothing (weighting the current poll as 3, the previous as 2 and the one before as 1 at any time).

Looking at the pattern between changes in the uncommitted vote and changes in the Labor 2PP since April 2011, the graph looks like this:


The unsmoothed version didn't look that different.  I also tried smoothing each poll using the polls both before and after it to see if that made things any clearer; it didn't.

It's pointing in the direction that it would be expected, but even as a one-tailed relationship it isn't quite statistically significant, and even if it was, it would explain a feeble six percent of variation in 2PP from poll to poll, so it wouldn't be worth getting excited about anyway.  Even if it was assumed that this relationship is real and valid, the degree of narrowing it would project between now and the final Newspoll would be substantial but not massive: about two points.  (A point of decline in the upfront exclusion rate would then be worth 0.37 points to the Labor 2PP).

But I think using such findings to try to project eventual narrowing would be unsound reasoning anyway.  Quite aside from the usual reservation that correlation isn't necessarily causation, it's one thing to argue that when the government does popular things it causes undecided voters to move into its camp, and when the government does unpopular things it causes its voters to become undecided.  It makes sense that that would be the case, even if there might be too much noise to prove it.  It's another to argue that as the approach of the election causes undecided voters to develop a voting intention, that they will move into the government's camp whether the government does especially popular things or not.

The view that there could be some narrowing in the government's favour is a reasonable (but far from necessarily true) position which is compatible with historical data and with suggestions of a possible relationship between the Government's struggles and the Newspoll upfront refusal rate.  But the view that the high uncommitted vote could save the Government entirely, and that it creates genuine doubt about the election outcome just doesn't show any signs of being justified.

No comments:

Post a Comment